arXiv subjects

Aaron D. Kaplan

Publications and source records attributed to Aaron D. Kaplan.

At least 19 recordsLinked to original sources

arXiv2025

A foundation model for atomistic materials chemistry

Atomistic simulations of matter, especially those that leverage first-principles (ab initio) electronic structure theory, provide a microscopic view of the world, underpinning much of our understanding of chemistry and materials science. Over the last decade or so, machine-learned force fields have transformed atomistic modeling by enabling simulations of ab initio quality over unprecedented time and length scales. However, early ML force fields have largely been limited by: (i) the substantial computational and human effort of developing and validating potentials for each particular system of interest; and (ii) a general lack of transferability from one chemical system to the next. Here we show that it is possible to create a general-purpose atomistic ML model, trained on a public dataset of moderate size, that is capable of running stable molecular dynamics for a wide range of molecules and materials. We demonstrate the power of the MACE-MP-0 model - and its qualitative and at times quantitative accuracy - on a diverse set of problems in the physical sciences, including properties of solids, liquids, gases, chemical reactions, interfaces and even the dynamics of a small protein. The model can be applied out of the box as a starting or "foundation" model for any atomistic system of interest and, when desired, can be fine-tuned on just a handful of application-specific data points to reach ab initio accuracy. Establishing that a stable force-field model can cover almost all materials changes atomistic modeling in a fundamental way: experienced users get reliable results much faster, and beginners face a lower barrier to entry. Foundation models thus represent a step towards democratising the revolution in atomic-scale modeling that has been brought about by ML force fields.

physics.chem-ph↗

arXiv2025

MP-ALOE: An r2SCAN dataset for universal machine learning interatomic potentials

We present MP-ALOE, a dataset of nearly 1 million DFT calculations using the accurate r2SCAN meta-generalized gradient approximation. Covering 89 elements, MP-ALOE was created using active learning and primarily consists of off-equilibrium structures. We benchmark a machine learning interatomic potential trained on MP-ALOE, and evaluate its performance on a series of benchmarks, including predicting the thermochemical properties of equilibrium structures; predicting forces of far-from-equilibrium structures; maintaining physical soundness under static extreme deformations; and molecular dynamic stability under extreme temperatures and pressures. MP-ALOE shows strong performance on all of these benchmarks, and is made public for the broader community to utilize.

cond-mat.mtrl-sci↗

arXiv2025

The Interplay Between Electron Localization, Magnetic Order, and Jahn-Teller Distortion that Dictates LiMnO$_2$ Phase Stability

The development of Mn-rich cathodes for Li-ion batteries promises to alleviate supply chain bottlenecks in battery manufacturing. Challenges in Mn-rich cathodes arise from Jahn-Teller (JT) distortions of Mn$^{3+}$, Mn migration, and phase transformations to spinel-like order, which can affect the electrochemical performance. These phenomena motivate an ab initio re-examination of the thermodynamics of the LiMnO2 polymorphs. It is found that the generalized gradient approximation (GGA - PBEsol) and meta-GGA (r2SCAN) density functionals with empirical on-site Hubbard U corrections yield spurious stable phases for LiMnO2, such as predicting a phase with gamma-LiFeO2-like order (g-LiMnO2) to be the ground state instead of the orthorhombic (Pmmn) phase, which is the experimentally known ground state. Accounting for the antiferromagnetic (AFM) order in each structure has a substantial effect on the total energies and resulting phase stability. By using hybrid-GGA (HSE06) and GGA with self-consistent Hubbard parameters (on-site U and inter-site V), the experimentally observed LiMnO2 phase stability trends are recovered. The calculated Hubbard U in the experimentally observed orthorhombic, layered, and spinel phases are significantly smaller than U in g-LiMnO2 and disordered layered structures. The smaller values of U are correlated with a collinear ordering of JT distortions, in which all $e_g$ orbitals are oriented in the same direction. This cooperative JT effect leads to increased Mn-O covalency, which contributes to the greater electronic stability compared to the phases with noncollinear JT arrangements, and also generate greater vibrational entropy, which helps stabilize these phases at high temperature. These phases are shown to be strongly insulating with large calculated band gaps > 3 eV, computed using HSE06 and $G_0W_0$.

cond-mat.mtrl-sci↗

arXiv2025

Accelerated discovery of cost-effective photoabsorber materials for near-infrared (λ=1600 nm) photodetector applications

Current infrared sensing devices are based on costly materials with relatively few viable alternatives known. To identify promising candidate materials for infrared photodetection, we have developed a high-throughput screening methodology based on high-accuracy r$^2$SCAN and HSE calculations in density functional theory. Using this method, we identify ten already synthesized materials between the inverse perovskite family, barium silver pnictide family, the alkaline pnictide family, and ZnSnAs$_2$ as top candidates. Among these, ZnSnAs$_2$ emerges as the most promising candidate due to its experimentally verified band gap of 0.74 eV at 0 K, and its cost-effective synthesis through Bridgman growth. BaAgP also shows potential with an HSE-calculated band gap of 0.64 eV, although further experimental validation is required. Lastly, we discover an additional material, Ca$_3$BiP, which has not been previously synthesized, but exhibits a promising optical spectra and a band gap of 0.56 eV. The method applied in this work is sufficiently general to screen wider bandgap materials in high-throughput and now extended to narrow-band gap materials.

cond-mat.mtrl-sci↗

arXiv2025

Towards High-Voltage Cathodes for Zinc-Ion Batteries: Discovery Pipeline and Material Design Rules

Efficient energy storage systems are crucial to address the intermittency of renewable energy sources. As multivalent batteries, Zn-ion batteries (ZIBs), while inherently low voltage, offer a promising low cost alternative to Li-ion batteries due to viable use of zinc as the anode. However, to maximize the potential impact of ZIBs, rechargable cathodes with improved Zn diffusion are needed. To better understand the chemical and structural factors influencing Zn-ion mobility within battery electrode materials, we employ a high-throughput computational screening approach to systematically evaluate candidate intercalation hosts for ZIB cathodes, expanding the chemical search space on empty intercalation hosts that do not contain Zn. We leverage a high-throughput screening funnel to identify promising cathodes in ZIBs, integrating screening criteria with DFT-based calculations of Zn$^{2+}$ intercalation and diffusion inside the host materials. Using this data, we identify the design principles that favor Zn-ion mobility in candidate cathode materials. Building on previous work on divalent ion cathodes, this study broadens the chemical space for next-generation multivalent energy storage systems.

cond-mat.mtrl-sci↗

arXiv2025

Cross-functional transferability in universal machine learning interatomic potentials

The rapid development of universal machine learning interatomic potentials (uMLIPs) has demonstrated the possibility for generalizable learning of the universal potential energy surface. In principle, the accuracy of uMLIPs can be further improved by bridging the model from lower-fidelity datasets to high-fidelity ones. In this work, we analyze the challenge of this transfer learning problem within the CHGNet framework. We show that significant energy scale shifts and poor correlations between GGA and r$^2$SCAN pose challenges to cross-functional data transferability in uMLIPs. By benchmarking different transfer learning approaches on the MP-r$^2$SCAN dataset of 0.24 million structures, we demonstrate the importance of elemental energy referencing in the transfer learning of uMLIPs. By comparing the scaling law with and without the pre-training on a low-fidelity dataset, we show that significant data efficiency can still be achieved through transfer learning, even with a target dataset of sub-million structures. We highlight the importance of proper transfer learning and multi-fidelity learning in creating next-generation uMLIPs on high-fidelity data.

cond-mat.mtrl-sci↗

arXiv2025

A Foundational Potential Energy Surface Dataset for Materials

Accurate potential energy surface (PES) descriptions are essential for atomistic simulations of materials. Universal machine learning interatomic potentials (UMLIPs)$^{1-3}$ offer a computationally efficient alternative to density functional theory (DFT)$^4$ for PES modeling across the periodic table. However, their accuracy today is fundamentally constrained due to a reliance on DFT relaxation data.$^{5,6}$ Here, we introduce MatPES, a foundational PES dataset comprising $\sim 400,000$ structures carefully sampled from 281 million molecular dynamics snapshots that span 16 billion atomic environments. We demonstrate that UMLIPs trained on the modestly sized MatPES dataset can rival, or even outperform, prior models trained on much larger datasets across a broad range of equilibrium, near-equilibrium, and molecular dynamics property benchmarks. We also introduce the first high-fidelity PES dataset based on the revised regularized strongly constrained and appropriately normed (r$^2$SCAN) functional$^7$ with greatly improved descriptions of interatomic bonding. The open source MatPES initiative emphasizes the importance of data quality over quantity in materials science and enables broad community-driven advancements toward more reliable, generalizable, and efficient UMLIPs for large-scale materials discovery and design.

cond-mat.mtrl-sci↗

arXiv2024

Realistic non-collinear ground states of solids with source-free exchange correlation functional

In this work, we extend the source-free (SF) exchange correlation (XC) functional developed by Sangeeta Sharma and co-workers to plane-wave density functional theory (DFT) based on the projector augmented wave (PAW) method. This constraint is implemented by the current authors within the VASP source code, using a fast Poisson solver that capitalizes on the parallel three-dimensional fast Fourier transforms (FFTs) implemented in VASP. Using this modified XC functional, we explore the improved convergence behavior that results from applying this constraint to the GGA-PBE+$U$+$J$ functional. In the process, we compare the non-collinear magnetic ground state computed by each functional and their SF counterpart for a select number of magnetic materials in order to provide a metric for comparing with experimentally determined magnetic orderings. We observe significantly improved agreement with experimentally measured magnetic ground state structures after applying the source-free constraint. Furthermore, we explore the importance of considering probability current densities in spin polarized systems, even under no applied field. We analyze the XC torque as well, in order to provide theoretical and computational analyses of the net XC magnetic torque induced by the source-free constraint. Along these lines, we highlight the importance of properly considering the real-space integral of the source-free local magnetic XC field. Our analysis on probability currents, net torque, and constant terms draws additional links to the rich body of previous research on spin-current density functional theory (SCDFT), and paves the way for future extensions and corrections to the SF corrected XC functional.

cond-mat.mtrl-sci↗

arXiv2023

Unconventional Error Cancellation Explains the Success of Hartree-Fock Density Functional Theory for Barrier Heights

Energy barriers, which control the rates of chemical reactions, are seriously underestimated by computationally-efficient semi-local approximations for the exchange-correlation energy. The accuracy of a semi-local density functional approximation is strongly boosted for reaction barrier heights by evaluating that approximation non-self-consistently on Hartree-Fock electron densities, as known for about 30 years. The conventional explanation is that Hartree-Fock theory yields the more accurate density. This article presents a benchmark Kohn-Sham inversion of accurate coupled-cluster densities for the reaction H$_2$ + F $\rightarrow$ HHF $\rightarrow$ H + HF, and finds a strong, understandable cancellation between positive (excessively over-corrected) density-driven and large negative functional-driven errors (expected from stretched radical bonds in the transition state) within this Hartree-Fock density functional theory. This confirms earlier conclusions [Kaplan et al., J. Chem. Theory Comput. 19, 532--543 (2023)] based on 76 barrier heights and three less reliable, but less expensive, fully-nonlocal density-functional proxies for the exact density.

physics.chem-ph↗

arXiv2023

Revealing quasi-excitations in the low-density homogeneous electron gas with model exchange-correlation kernels

Time-dependent density functional theory (TDDFT) within the linear response regime provides a solid mathematical framework to capture excitations. The accuracy of the theory, however, largely depends on the approximations for the exchange-correlation (xc) kernels. Away from the long-wavelength (or $q=0$ short wave-vector) and zero-frequency ($ω=0$) limit, the correlation contribution to the kernel becomes more relevant and dominant over exchange. The dielectric function in principle can encompass xc effects relevant to describe low-density physics. Furthermore, besides collective plasmon excitations, the dielectric function can reveal collective electron-hole excitations, often dubbed ``ghost excitons.'' Beside collective excitons, the physics in the low-density regime is rich, as exemplified by a static charge-density wave that was recently found for $r_\mathrm{s} > 69$, and was shown to be associated with softening of the plasmon mode. These excitations are seen to be present in much higher density 2D HEGs, of $r_\mathrm{s} \geq 4$. In this work we perform a thorough analysis with xc model kernels for excitations of various nature. The uniform electron gas as a useful model of real metallic systems is used as a platform of our analysis. We highlight the relevance of exact constraints as we display and explain screening and excitations in the low-density region.

cond-mat.str-el↗

arXiv2023

QMC-consistent static spin and density local field factors for the uniform electron gas

Analytic mathematical models for the static spin ($G_-$) and density ($G_+$) local field factors for the uniform electron gas (UEG) as functions of wavevector and density are presented. These models closely fit recent quantum Monte Carlo (QMC) data and satisfy exact asymptotic limits. This model for $G_-$ is available for the first time, and the present model for $G_+$ is an improvement over previous work. The QMC-computed $G_\pm$ are consistent with a rapid crossover between theoretically-derived small-$q$ and large-$q$ expansions of $G_\pm$. These expansions are completely determined by $r_\mathrm{s}$, the UEG correlation energy per electron, and the UEG on-top pair distribution function. We demonstrate their utility by computing uniform electron gas correlation energies over a range of densities. These models, which hold over an extremely wide range of densities, are recommended for use in practical time-dependent density functional theory calculations of simple metallic systems. A revised model of the spin susceptibility enhancement is developed that fits QMC data, and does not show a ferromagnetic instability at low density.

cond-mat.str-el↗

arXiv2022

Understanding density driven errors for reaction barrier heights

Delocalization errors, such as charge-transfer and some self-interaction errors, plague computationally-efficient and otherwise-accurate density functional approximations (DFAs). Evaluating a semi-local DFA non-self-consistently on the Hartree-Fock (HF) density is often recommended as a computationally cheap remedy for delocalization errors. For sophisticated meta-GGAs like SCAN, this approach can achieve remarkable accuracy. When this HF-DFT (or DFA@HF) significantly improves over the DFA, it is often presumed that the HF density is more accurate than the self-consistent DFA density. By applying the metrics of density-corrected density functional theory (DFT), we show that HF-DFT works for barrier heights by making a localizing charge transfer error or density over-correction, thereby producing a somewhat-reliable cancellation of density- and functional-driven errors for the energy. A quantitative analysis of the charge transfer errors in a few transition states confirms this trend. We do not have the exact functional and exact densities that are needed to evaluate the exact density- and functional-driven errors for the large BH76 database of barrier heights. Instead, we have identified and used three non-local proxy functionals (the SCAN 50% global hybrid, the range-separated hybrid LC-$ω$PBE, and SCAN-FLOSIC) and their self-consistent densities. These functionals yield reasonably accurate self-consistent barrier heights, and their self-consistent total energies are nearly piecewise linear in fractional electron number - two important points of similarity to the exact functional. We argue that density-driven errors of the energy in a self-consistent density functional calculation are second-order in the density error, and that large density-driven errors arise primarily from incorrect electron transfers over length scales larger than the diameter of an atom.

physics.chem-ph↗

arXiv2022

Symmetry Breaking with the SCAN Density Functional Describes Strong Correlation in the Singlet Carbon Dimer

The SCAN (strongly constrained and appropriately normed) meta-generalized gradient approximation (meta-GGA), which satisfies all 17 exact constraints that a meta-GGA can satisfy, accurately describes equilibrium bonds that are normally correlated. With symmetry breaking, it also accurately describes some sd equilibrium bonds that are strongly correlated. While sp equilibrium bonds are nearly always normally correlated, the C2 singlet ground state is known to be a rare case of strong correlation in an sp equilibrium bond. Earlier work that calculated atomization energies of the molecular sequence B2, C2, O2, and F2 in the local spin density approximation (LSDA), the Perdew-Burke-Ernzerhof (PBE) GGA, and the SCAN meta-GGA, without symmetry breaking in the molecule, found that only SCAN was accurate enough to reveal an anomalous under-binding for C2. This work shows that spin symmetry breaking in singlet C2, the appearance of net up- and down-spin densities on opposite sides (not ends) of the bond, corrects that under-binding, with a small SCAN atomization-energy error more like that of the other three molecules, suggesting that symmetry-breaking with an advanced density functional might reliably describe strong correlation. This article also discusses some general aspects of symmetry breaking, and the insights into strong correlation that symmetry-breaking can bring.

physics.chem-ph↗

arXiv2022

Testing the r$^2$SCAN density functional for the thermodynamic stability of solids with and without a van der Waals correction

A central aim of materials discovery is an accurate and numerically reliable description of thermodynamic properties, such as the enthalpies of formation and decomposition. The r$^2$SCAN revision of the strongly constrained and appropriately normed (SCAN) meta-generalized gradient approximation (meta-GGA) balances numerical stability with high general accuracy. To assess the r$^2$SCAN description of solid-state thermodynamics, we evaluate the formation and decomposition enthalpies, equilibrium volumes, and fundamental bandgaps of more than 1,000 solids using r$^2$SCAN, SCAN, and PBE, as well as two dispersion-corrected variants, SCAN+rVV10 and r$^2$SCAN+rVV10. We show that r$^2$SCAN achieves accuracy comparable to SCAN and often improves upon SCAN's already excellent accuracy. Whereas SCAN+rVV10 is often observed to worsen the formation enthalpies of SCAN, and makes no substantial correction to SCAN's cell volume predictions, r$^2$SCAN+rVV10 predicts marginally less-accurate formation enthalpies than r$^2$SCAN, and slightly more-accurate cell volumes than r$^2$SCAN. The average absolute errors in predicted formation enthalpies are found to decrease by a factor of 1.5 to 2.5 from the GGA level to the meta-GGA level. Smaller decreases in error are observed for decomposition enthalpies. For formation enthalpies r$^2$SCAN improves over SCAN for intermetallic systems. For a few classes of systems -- transition metals, intermetallics, weakly-bound solids, and enthalpies of decomposition into compounds -- GGAs are comparable to meta-GGAs. In total, r$^2$SCAN and r$^2$SCAN+rVV10 can be recommended as stable, general-purpose meta-GGAs for materials discovery.

cond-mat.mtrl-sci↗

arXiv2022

Laplacian-level meta-generalized gradient approximation for solid and liquid metals

We derive and motivate a Laplacian-level, orbital-free meta-generalized-gradient approximation (LL-MGGA) for the exchange-correlation energy, targeting accurate ground-state properties of $sp$ and $sd$ metallic condensed matter, in which the density functional for the exchange-correlation energy is only weakly nonlocal due to perfect long-range screening. Our model for the orbital-free kinetic energy density restores the fourth-order gradient expansion for exchange to the r$^2$SCAN meta-GGA [Furness et al., J. Phys. Chem. Lett. 11, 8208 (2020)], yielding a LL-MGGA we call OFR2. OFR2 matches the accuracy of SCAN for prediction of common lattice constants and improves the equilibrium properties of alkali metals, transition metals, and intermetallics that were degraded relative to the PBE GGA values by both SCAN and r$^2$SCAN. We compare OFR2 to the r$^2$SCAN-L LL-MGGA [D. Mejia-Rodriguez and S.B. Trickey, Phys. Rev. B 102, 121109 (2020)] and show that OFR2 tends to outperform r$^2$SCAN-L for the equilibrium properties of solids, but r$^2$SCAN-L much better describes the atomization energies of molecules than OFR2 does. For best accuracy in molecules and non-metallic condensed matter, we continue to recommend SCAN and r$^2$SCAN. Numerical performance is discussed in detail, and our work provides an outlook to machine learning.

cond-mat.mtrl-sci↗

arXiv2022

Workhorse minimally-empirical dispersion-corrected density functional, with tests for weakly-bound systems: r$^{2}$SCAN+rVV10

SCAN+rVV10 has been demonstrated to be a versatile van der Waals (vdW) density functional that delivers good predictions of both energetic and structural properties for many types of bonding. Recently, the r$^{2}$SCAN functional has been devised as a revised form of SCAN with improved numerical stability. In this work, we refit the rVV10 functional to optimize the r$^{2}$SCAN+rVV10 vdW density functional, and test its performance for molecular interactions and layered materials. Our molecular tests demonstrate that r$^{2}$SCAN+rVV10 outperforms its predecessor SCAN+rVV10 in both efficiency (numerical stability) and accuracy. This good performance is also found in lattice constant predictions. In comparison with benchmark results from higher-level theories or experiments, r$^{2}$SCAN+rVV10 yields excellent interlayer binding energies and phonon dispersions for layered materials.

cond-mat.mtrl-sci↗

arXiv2022

Predictive Power of the Exact Constraints and Appropriate Norms in Density Functional Theory

Ground-state Kohn-Sham density functional theory provides, in principle, the exact ground-state energy and electronic spin-densities of real interacting electrons in a static external potential. In practice, the exact density functional for the exchange-correlation (xc) energy must be approximated in a computationally efficient way. About twenty mathematical properties of the exact xc functional are known. In this work, we review and discuss these known constraints on the xc energy and hole. By analyzing a sequence of increasingly sophisticated density functional approximations (DFAs), we argue that: (1) the satisfaction of more exact constraints and appropriate norms makes a functional more predictive over the immense space of many-electron systems; (2) fitting to bonded systems yields an interpolative DFA that may not extrapolate well to systems unlike those in the fitting set. We discuss how the class of well-described systems has grown along with constraint satisfaction, and the possibilities for future functional development.

physics.chem-ph↗

arXiv2021

First-principles wavevector- and frequency-dependent exchange-correlation kernel for jellium at all densities

We propose a spatially and temporally nonlocal exchange-correlation (xc) kernel for the spin-unpolarized fluid phase of ground-state jellium, for use in time-dependent density functional and linear response calculations. The kernel is constructed to satisfy known properties of the exact xc kernel, to accurately describe the correlation energies of bulk jellium, and to satisfy frequency-moment sum rules at a wide range of bulk jellium densities, including those low densities that display strong correlation and symmetry breaking. These effects are easier to understand in the simple jellium model than in real systems. All exact constraints satisfied by the recent MCP07 kernel [A. Ruzsinszky, et al., Phys. Rev. B 101, 245135 (2020)] are maintained in the new revised MCP07 (rMCP07) kernel, while others are added. The revision $f_\mathrm{xc}^\mathrm{rMCP07}(q,ω)$ differs from MCP07 only for non-zero frequencies $ω$. Only at densities much lower than those of real bulk metals is the frequency dependence of the kernel important for the correlation energy of jellium. As the wavevector $q$ tends to zero, the kernel has a $-4πα(ω)/q^2$ divergence whose frequency-dependent ultranonlocality coefficient $α(ω)$ vanishes in jellium, and is predicted by rMCP07 to be extremely small for the real metals Al and Na.}

cond-mat.mtrl-sci↗